Single tool 'run_process' has a well-structured schema with proper type definitions and enum constraints for the 'mode' parameter. Description is adequate (67 chars) but generic. All parameters have descriptions and types. However, the tool lacks critical safety guardrails: no dry-run validation guidance, no error recovery hints, and no output schema documentation. The tool is destructive (can execute arbitrary commands) yet lacks confirmation/audit patterns. Parameter descriptions are present but could be more actionable regarding constraints and error cases.

• Document the output schema in the tool description or as a structured example: 'Returns {stdout: string, stderr: string, exit_code: number, timed_out: boolean}'.
• Add explicit warning to description: 'WARNING: This tool can execute destructive commands. Always use dry_run=true first to preview what will execute. Irreversible operations (rm, dd, etc.) cannot be undone.'
• Enhance 'command_line' parameter description with escaping guidance: 'Shell command line (bash/sh syntax). Use quotes for arguments with spaces. Avoid unescaped special chars ($, `, |, &, ;, >, <).'
• Add error recovery hints: 'If timeout occurs, increase timeout_ms or break the command into smaller steps. If permission denied, verify the user running this server has required access.'
• Document the security model: 'Executes commands as the user running the MCP server. No sandboxing. Requires explicit trust in the LLM/agent calling this tool.'